french-education-speech / merge_transcripts.py
mathisescriva
Initial commit: French Education Speech Corpus (Phase 1)
5d2c0a9
#!/usr/bin/env python3
import argparse
import json
from pathlib import Path
from typing import Dict
import orjson
def load_jsonl(path: Path):
with open(path, "r", encoding="utf-8") as f:
for line in f:
if line.strip():
yield json.loads(line)
def main() -> None:
parser = argparse.ArgumentParser(description="Merge hypothesis texts into base JSONL by id.")
parser.add_argument("--base", required=True, help="Base JSONL with entries containing id and audio...")
parser.add_argument("--hyp", required=True, help="Hypothesis JSONL with id,text")
parser.add_argument("--out", required=True, help="Output JSONL (overwrites base or writes new)")
args = parser.parse_args()
hyp_map: Dict[str, str] = {}
for rec in load_jsonl(Path(args.hyp)):
txt = (rec.get("text") or "").strip()
if txt:
hyp_map[rec["id"]] = txt
with open(args.out, "w", encoding="utf-8") as out_f:
for base in load_jsonl(Path(args.base)):
_id = base["id"]
if _id in hyp_map:
base["text"] = hyp_map[_id]
out_f.write(orjson.dumps(base).decode("utf-8") + "\n")
print(f"Merged transcripts -> {args.out} (updated {len(hyp_map)} items)")
if __name__ == "__main__":
main()